Mid-term Forecast of Electricity Consumption of Enterprises Based on Bi-Directional LSTM and Nonpooling CNN

Abstract

Mid-term forecast of the electricity consumption of enterprises (MFECE) plays a critical role in enterprises' power planning, economical operation, and energy management. This article proposes a novel hybrid neural network-based forecast scheme to reduce the uncertainty and instability of MFECE. The proposed hybrid neural network consists of the bi-directional long short-term memory (BLSTM) network and the nonpooling convolutional neural network (NPCNN). This model uses electricity load data and external features such as calendar, weather, and holiday information as input. NPCNN is used to extract features from the input data set. Then the extracted features and electricity load data are input into BLSTM for training, and the electric load is predicted. The proposed method is tested on a set of real enterprise load data sets and compared with several classical algorithms on the same data sets. The experimental results have proved its superiority.